Modeling Power Consumption in Graphics Processors

Abstract

A graphics processor (GPU) is a dedicated processing unit that handles all graphics-related computation in a given system. With the growth of graphics-based applications in domains such as embedded systems, gaming, and image processing, studying such systems has become more important. We propose to develop a simulator that can be used to analyze energy efficiency for GPUs. No publicly available simulators simulate power utilization in GPUs, which is important because energy consumption is one of the key design criteria for any modern system. Existing power simulators like Simple Power, Wattch, and PowerTimer are not GPU specific. GPU simulators, such as ATTILA, that simulate the graphics pipeline thoroughly do not model power. Our plan is to use the existing ATTILA simulator and extend it to do the power modeling. This thesis focuses on modeling GPU caches using a combination of ATTILA and CACTI, a widely used cache power estimation tool.